Runway Characters adds custom voices from text prompts with API access

• Runway's launch post says Custom Voices for Characters can now be generated from text prompts, so creators can specify tone, pacing, accent, and personality instead of starting from a recorded source voice.
• According to Runway's follow-up post, the feature is live in both the web app and the API, with the docs pointing to a new custom voices reference.
• The Characters overview frames voice as one of the core controls alongside personality, knowledge, and actions, which makes this a workflow upgrade for branded avatars and animated hosts.
• Runway co-founder Cristóbal Valenzuela called custom voices the most requested feature for real-time avatars, which matches how central voice setup has become in the Characters product.

The new docs quietly add a second path too: you can still clone from an audio sample, and both prompt-made and cloned voices end up in the same Voices API. The Characters guide also shows voice selection already sitting inside the character creation flow, so this update plugs directly into a product creators were already using for live avatar calls.

The headline change is simple: Runway Characters can now get a voice from a text description. In the API docs, the prompt-based flow asks for a name, a text description, and a model choice, with eleven_ttv_v3 listed as the latest option.

The prompt itself is supposed to describe voice traits, not a script. The docs call out tone, accent, pacing, and personality, and Runway's demo shows that workflow turning a prompt like "a confident, warm male voice" into multiple generated options.

Runway did not ship this as an API-only feature. The announcement thread says it is available in the web app now, while the Create Your Own Characters tutorial places voice choice directly inside the character builder.

That matters for creative teams working across prototyping and production. The same Characters stack already supports a single-image avatar with configurable voice, personality, knowledge, and actions, according to the product docs.

The docs make clear this is an addition, not a replacement. The custom voice page keeps voice cloning alongside text-based voice design, with audio samples accepted from 10 seconds to 5 minutes and capped at 10 MB.

It also exposes a few implementation details that are easy to miss in the launch posts:

• voice creation is asynchronous
• new voices move through PROCESSING, READY, or FAILED
• a previewUrl appears when the voice is ready
• the finished voice can be assigned to any Avatar as a custom voice
• organizations can list and manage voices through the same Voices API and Developer Portal